Loading organizations...

§ Private Profile · Cupertino, CA, USA
AI semiconductor company designing custom ASICs for large language model inference, focused on transformer AI architecture.
Based in Cupertino, California, Etched.ai is a semiconductor startup that designs custom application-specific integrated circuits specifically optimized for artificial intelligence workloads. The company's flagship product, Sohu, is a specialized chip that embeds transformer models directly into hardware to process large language models with significantly greater speed and lower inference costs than general-purpose graphics processing units. Operating with a team of approximately 35 employees, the pre-revenue hardware developer fabricates its specialized processors through a strategic manufacturing partnership with TSMC. Etched.ai has secured nearly $1 billion in total funding, including a recent $500 million financing round that valued the enterprise at approximately $5 billion, drawing capital from prominent venture investors such as Peter Thiel, Stripes, and Primary Venture Partners. The semiconductor company was founded in 2022 by Gavin Uberti, Chris Zhu, and Robert Wachen.
Etched.ai has raised $625.0M across 3 funding rounds.
Etched.ai has raised $625.0M in total across 3 funding rounds.
Etched.ai has raised $625.0M in total across 3 funding rounds.
Etched.ai's investors include Stripes, Peter Thiel, Positive Sum, Ribbit Capital, Primary Venture Partners, Accel, AngelList Syndicator, ARCH Venture Partners, C2 Investment, Cedar Capital Group, Coatue, Craft Ventures.
Etched.ai is a semiconductor startup building specialized ASICs for AI inference, with its flagship product Sohu, the first chip purpose-built for transformer models that power large language models like ChatGPT, Claude, and Gemini.[2][3][5] It serves AI developers and enterprises running demanding inference workloads, such as real-time voice agents, content generation, and trillion-parameter models, solving the problem of GPU limitations in speed, cost, and power efficiency by delivering over 500,000 tokens per second on Llama-70B with an 8x Sohu server—outperforming 8x Nvidia H100 GPUs by more than 20x.[1][2][3] Founded in 2022 and based in Cupertino, California, Etched has raised $125 million, including a $120 million Series A in June 2024, fueling rapid growth toward challenging Nvidia's dominance in AI hardware.[2][4][5]
Etched was founded in 2022 by Harvard dropouts Gavin Uberti (ex-OctoML, ex-Xnor.ai), Chris Zhu, and Robert Wachen, who bonded over a school project exploring AI at scale and realized transformers would dominate future AI.[2][4][5] Convinced general-purpose GPUs couldn't keep pace, they dropped out to build specialized chips, starting from a dorm-room vision of outperforming Nvidia for inference on transformer architectures like those in ChatGPT and Stable Diffusion.[2][4] Early traction came via a $5.4 million seed round in March 2023 from angel investors, followed by the June 2024 Series A co-led by Primary Venture Partners and Positive Sum Ventures, with backers like Peter Thiel, GitHub CEO Thomas Dohmke, and Cruise co-founder Kyle Vogt—validating their bold pivot and enabling TSMC fabrication of Sohu.[2][5]
Etched stands out by hardwiring the transformer architecture directly into silicon, eliminating GPU overhead for non-transformer tasks:
Collaborations like Rambus have minimized chip area and accelerated time-to-market.[4]
Etched rides the transformer dominance trend—architectures behind 90%+ of generative AI since Google's 2017 paper, fueling models from OpenAI's Sora to Stable Diffusion amid exploding inference demand.[2][5] Timing is ideal post-ChatGPT boom, as GPU shortages and Nvidia's pricing/power issues create openings for ASICs; market forces like hyperscaler needs for real-time AI (e.g., agents, video gen) favor specialists over generalists.[1][3][5] Etched influences the ecosystem by open-sourcing tools, enabling startups to run trillion-param models affordably, and pressuring incumbents—positioning as a Nvidia challenger in a $100B+ AI chip race with competitors like Recogni, Together AI, and MatX.[1][2]
Etched is poised to ship Sohu servers in 2025-2026, scaling production via TSMC and targeting hyperscalers for trillion-param inference as AI shifts from training to ubiquitous deployment.[2][3] Trends like agentic AI, multimodal models, and edge inference will amplify demand for its efficiency, potentially capturing share from Nvidia if software maturity holds. Influence may evolve from niche pioneer to ecosystem shaper, fostering open transformer hardware amid U.S.-China chip tensions—echoing its dorm-room origins as a bet on specialized silicon transforming AI from GPU bottleneck to real-time reality.[4][5]
Etched.ai has raised $625.0M across 3 funding rounds. Most recently, it raised $500.0M Other Equity in January 2026.